Background and motivation

Integrating terrestrial cellular networks with non-terrestrial networks, such as Low Earth Orbit (LEO) satellite systems, can improve IoT connectivity in areas with intermittent or unreliable terrestrial coverage. However, switching between terrestrial and satellite links introduces challenges due to changing link availability, communication delays, and switching overhead. Predicting link conditions may improve connectivity decisions, but inaccurate predictions can also result in unnecessary switching or service interruptions.

Objective

This thesis will investigate whether uncertainty-aware prediction of link availability can improve IoT service continuity compared with conventional reactive and prediction-only link-selection strategies.

Proposed work

The work will develop or identify a suitable existing simulation environment representing an IoT device with access to a terrestrial cellular link and a satellite link. A lightweight probabilistic machine-learning model, such as logistic regression or gradient-boosted trees, will estimate short-term link availability using representative connectivity traces. The predicted probabilities will be calibrated and incorporated into a simple confidence-guarded link-switching policy.

The predictive link-selection strategy will account for both forecast information and its uncertainty. When prediction confidence is insufficient, the controller may retain the current link or fall back to a conservative switching policy. The approach will be compared with terrestrial-first selection, reactive switching, and predictive switching without uncertainty awareness. Evaluation will consider service availability, communication delay, service interruptions, switching frequency, and the effects of prediction errors and switching overhead.

Expected outcomes

A TN–NTN simulation framework, a lightweight uncertainty-aware predictive switching algorithm, and a comparative assessment of its performance under changing connectivity conditions.

Contact

Fazal Khan <fazal.khan@kau.se>, Giuseppe Caso <giuseppe.caso@kau.se>, Anna Brunstrom <anna.brunstrom@kau.se>